Artificial Intelligence Nanodegree¶

Convolutional Neural Networks¶

Project: Write an Algorithm for a Dog Identification App¶


In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!

Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the iPython Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.

In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.

The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this IPython notebook.


Why We're Here¶

In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).

Sample Dog Output

In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!

The Road Ahead¶

We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.

  • Step 0: Import Datasets
  • Step 1: Detect Humans
  • Step 2: Detect Dogs
  • Step 3: Create a CNN to Classify Dog Breeds (from Scratch)
  • Step 4: Use a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 6: Write your Algorithm
  • Step 7: Test Your Algorithm

Step 0: Import Datasets¶

Import Dog Dataset¶

In the code cell below, we import a dataset of dog images. We populate a few variables through the use of the load_files function from the scikit-learn library:

  • train_files, valid_files, test_files - numpy arrays containing file paths to images
  • train_targets, valid_targets, test_targets - numpy arrays containing onehot-encoded classification labels
  • dog_names - list of string-valued dog breed names for translating labels
In [1]:
from sklearn.datasets import load_files       
  from keras.utils import np_utils
  import numpy as np
  from glob import glob
  
  # define function to load train, test, and validation datasets
  def load_dataset(path):
      data = load_files(path)
      dog_files = np.array(data['filenames'])
      dog_targets = np_utils.to_categorical(np.array(data['target']), 133)
      return dog_files, dog_targets
  
  # load train, test, and validation datasets
  train_files, train_targets = load_dataset('dogImages/train')
  valid_files, valid_targets = load_dataset('dogImages/valid')
  test_files, test_targets = load_dataset('dogImages/test')
  
  # load list of dog names
  dog_names = [item[20:-1] for item in sorted(glob("dogImages/train/*/"))]
  
  # print statistics about the dataset
  print('There are %d total dog categories.' % len(dog_names))
  print('There are %s total dog images.\n' % len(np.hstack([train_files, valid_files, test_files])))
  print('There are %d training dog images.' % len(train_files))
  print('There are %d validation dog images.' % len(valid_files))
  print('There are %d test dog images.'% len(test_files))
  
/usr/local/lib/python3.6/dist-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
    from ._conv import register_converters as _register_converters
  Using TensorFlow backend.
  
There are 133 total dog categories.
  There are 8351 total dog images.
  
  There are 6680 training dog images.
  There are 835 validation dog images.
  There are 836 test dog images.
  

Import Human Dataset¶

In the code cell below, we import a dataset of human images, where the file paths are stored in the numpy array human_files.

In [2]:
import random
  random.seed(8675309)
  
  # load filenames in shuffled human dataset
  human_files = np.array(glob("lfw/*/*"))
  random.shuffle(human_files)
  
  # print statistics about the dataset
  print('There are %d total human images.' % len(human_files))
  
There are 13233 total human images.
  

Step 1: Detect Humans¶

We use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images. OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory.

In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.

In [3]:
import cv2                
  import matplotlib.pyplot as plt                        
  %matplotlib inline                               
  
  # extract pre-trained face detector
  face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt.xml')
  
  # load color (BGR) image
  img = cv2.imread(human_files[120])
  # convert BGR image to grayscale
  gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
  
  # find faces in image
  faces = face_cascade.detectMultiScale(gray)
  
  # print number of faces detected in the image
  print('Number of faces detected:', len(faces))
  
  # get bounding box for each detected face
  for (x,y,w,h) in faces:
      # add bounding box to color image
      cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
      
  # convert BGR image to RGB for plotting
  cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
  
  # display the image, along with bounding box
  plt.imshow(cv_rgb)
  plt.show()
  
Number of faces detected: 1
  

Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.

In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.

Write a Human Face Detector¶

We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.

In [4]:
# returns "True" if face is detected in image stored at img_path
  def face_detector(img_path):
      img = cv2.imread(img_path)
      gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
      faces = face_cascade.detectMultiScale(gray)
      return len(faces) > 0
  

(IMPLEMENTATION) Assess the Human Face Detector¶

Question 1: Use the code cell below to test the performance of the face_detector function.

  • What percentage of the first 100 images in human_files have a detected human face?
  • What percentage of the first 100 images in dog_files have a detected human face?

Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.

Answer:

In [5]:
human_files_short = human_files[:100]
  dog_files_short = train_files[:100]
  # Do NOT modify the code above this line.
  
  ## TODO: Test the performance of the face_detector algorithm 
  ## on the images in human_files_short and dog_files_short.
  no_of_faces = [human_img for human_img in human_files_short if face_detector(human_img)]
  human_pc = len(no_of_faces) * 1.
  print("Human percentage: {}%".format(human_pc))
  
  no_of_dog = [dog_img for dog_img in dog_files_short if face_detector(dog_img)]
  dog_pc = len(no_of_dog) * 1.
  print("Dog percentage: {}%".format(dog_pc))
  
Human percentage: 97.0%
  Dog percentage: 12.0%
  

Question 2: This algorithmic choice necessitates that we communicate to the user that we accept human images only when they provide a clear view of a face (otherwise, we risk having unneccessarily frustrated users!). In your opinion, is this a reasonable expectation to pose on the user? If not, can you think of a way to detect humans in images that does not necessitate an image with a clearly presented face?

Answer:

We suggest the face detector from OpenCV as a potential way to detect human images in your algorithm, but you are free to explore other approaches, especially approaches that make use of deep learning :). Please use the code cell below to design and test your own face detection algorithm. If you decide to pursue this optional task, report performance on each of the datasets.

My answer:

I think that it's better to provide a clear view of a face is a good start.

To me, Haar cascades is a really handy and small facail detection library that even we can install in rasberryPI. In the other hand, I chose to use this dlib library is because I think moblie phone has a relatively bigger storage and its performance is better than Haar cascades which has an accuracy of 99.38% on the Labeled Faces in the Wild benchmark.

In [40]:
## (Optional) TODO: Report the performance of another  
  ## face detection algorithm on the LFW dataset
  ### Feel free to use as many code cells as needed.
  
  import face_recognition
  from PIL import Image, ImageDraw
  
  def draw_facial_features():
      image = face_recognition.load_image_file(human_files_short[0])
      face_landmarks_list = face_recognition.face_landmarks(image)
      print("Number of faces detected: {}".format(len(face_landmarks_list)))
  
      # Create a PIL imagedraw object so we can draw on the picture
      pil_image = Image.fromarray(image)
      d = ImageDraw.Draw(pil_image)
  
      for face_landmarks in face_landmarks_list:
  
          # Print the location of each facial feature in this image
          #for facial_feature in face_landmarks.keys():
          #    print("The {} in this face has the following points: {}".format(facial_feature, face_landmarks[facial_feature]))
  
          # Let's trace out each facial feature in the image with a line!
          for facial_feature in face_landmarks.keys():
              d.line(face_landmarks[facial_feature], width=5)
      display(pil_image)
  
  # Show the picture
  print("This is an example of using dib to do a Face Recognition function.")
  draw_facial_features()
  
  no_of_faces = [human_img for human_img in human_files_short if 
                 face_recognition.face_locations(face_recognition.load_image_file(human_img))
                ]
  human_pc = len(no_of_faces) * 1.
  print("\nSucessfully detected human face percentage: {}%".format(human_pc))
  
This is an example of using dib to do a Face Recognition function.
  Number of faces detected: 1
  
Sucessfully detected human face percentage: 100.0%
  

Step 2: Detect Dogs¶

In this section, we use a pre-trained ResNet-50 model to detect dogs in images. Our first line of code downloads the ResNet-50 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories. Given an image, this pre-trained ResNet-50 model returns a prediction (derived from the available categories in ImageNet) for the object that is contained in the image.

In [41]:
from keras.applications.resnet50 import ResNet50
  
  # define ResNet50 model
  ResNet50_model = ResNet50(weights='imagenet')
  

Pre-process the Data¶

When using TensorFlow as backend, Keras CNNs require a 4D array (which we'll also refer to as a 4D tensor) as input, with shape

(nb_samples,rows,columns,channels),

where nb_samples corresponds to the total number of images (or samples), and rows, columns, and channels correspond to the number of rows, columns, and channels for each image, respectively.

The path_to_tensor function below takes a string-valued file path to a color image as input and returns a 4D tensor suitable for supplying to a Keras CNN. The function first loads the image and resizes it to a square image that is 224×224 pixels. Next, the image is converted to an array, which is then resized to a 4D tensor. In this case, since we are working with color images, each image has three channels. Likewise, since we are processing a single image (or sample), the returned tensor will always have shape

(1,224,224,3).

The paths_to_tensor function takes a numpy array of string-valued image paths as input and returns a 4D tensor with shape

(nb_samples,224,224,3).

Here, nb_samples is the number of samples, or number of images, in the supplied array of image paths. It is best to think of nb_samples as the number of 3D tensors (where each 3D tensor corresponds to a different image) in your dataset!

In [42]:
from keras.preprocessing import image                  
  from tqdm import tqdm
  
  def path_to_tensor(img_path):
      
      # loads RGB image as PIL.Image.Image type
      img = image.load_img(img_path, target_size=(224, 224))
      
      # convert PIL.Image.Image type to 3D tensor with shape (224, 224, 3)
      x = image.img_to_array(img)
      
      # convert 3D tensor to 4D tensor with shape (1, 224, 224, 3) and return 4D tensor
      return np.expand_dims(x, axis=0)
  
  
  def paths_to_tensor(img_paths):
      
      list_of_tensors = [path_to_tensor(img_path) for img_path in tqdm(img_paths)]
      return np.vstack(list_of_tensors)
  

Making Predictions with ResNet-50¶

Getting the 4D tensor ready for ResNet-50, and for any other pre-trained model in Keras, requires some additional processing. First, the RGB image is converted to BGR by reordering the channels. All pre-trained models have the additional normalization step that the mean pixel (expressed in RGB as [103.939,116.779,123.68] and calculated from all pixels in all images in ImageNet) must be subtracted from every pixel in each image. This is implemented in the imported function preprocess_input. If you're curious, you can check the code for preprocess_input here.

Now that we have a way to format our image for supplying to ResNet-50, we are now ready to use the model to extract the predictions. This is accomplished with the predict method, which returns an array whose i-th entry is the model's predicted probability that the image belongs to the i-th ImageNet category. This is implemented in the ResNet50_predict_labels function below.

By taking the argmax of the predicted probability vector, we obtain an integer corresponding to the model's predicted object class, which we can identify with an object category through the use of this dictionary.

In [43]:
from keras.applications.resnet50 import preprocess_input, decode_predictions
  
  def ResNet50_predict_labels(img_path):
      # returns prediction vector for image located at img_path
      img = preprocess_input(path_to_tensor(img_path))
      return np.argmax(ResNet50_model.predict(img))
  

Write a Dog Detector¶

While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'. Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained ResNet-50 model, we need only check if the ResNet50_predict_labels function above returns a value between 151 and 268 (inclusive).

We use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).

In [44]:
### returns "True" if a dog is detected in the image stored at img_path
  # Sai: Only dictionary key between 151-268 are dog images!
  def dog_detector(img_path):
      prediction = ResNet50_predict_labels(img_path)
      return ((prediction <= 268) & (prediction >= 151))
  

(IMPLEMENTATION) Assess the Dog Detector¶

Question 3: Use the code cell below to test the performance of your dog_detector function.

  • What percentage of the images in human_files_short have a detected dog?
  • What percentage of the images in dog_files_short have a detected dog?

Answer:

In [11]:
### TODO: Test the performance of the dog_detector function
  ### on the images in human_files_short and dog_files_short.
  no_of_faces = [human_img for human_img in human_files_short if dog_detector(human_img)]
  human_pc = len(no_of_faces) * 1.
  print("Human percentage: {}%".format(human_pc))
  
  no_of_dog = [dog_img for dog_img in dog_files_short if dog_detector(dog_img)]
  dog_pc = len(no_of_dog) * 1.
  print("Dog percentage: {}%".format(dog_pc))
  
Human percentage: 0.0%
  Dog percentage: 100.0%
  

Step 3: Create a CNN to Classify Dog Breeds (from Scratch)¶

Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 1%. In Step 5 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.

Be careful with adding too many trainable layers! More parameters means longer training, which means you are more likely to need a GPU to accelerate the training process. Thankfully, Keras provides a handy estimate of the time that each epoch is likely to take; you can extrapolate this estimate to figure out how long it will take for your algorithm to train.

We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have great difficulty in distinguishing between a Brittany and a Welsh Springer Spaniel.

Brittany Welsh Springer Spaniel

It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).

Curly-Coated Retriever American Water Spaniel

Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.

Yellow Labrador Chocolate Labrador Black Labrador

We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.

Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!

Pre-process the Data¶

We rescale the images by dividing every pixel in every image by 255.

In [45]:
# Before processing image data
  def preprocess_data_display():
      tensor_images = paths_to_tensor(test_files[:1])
      display(tensor_images[0].shape)
      display(tensor_images[0][0])
  
  preprocess_data_display()
  
100%|██████████| 1/1 [00:00<00:00, 157.56it/s]
  
(224, 224, 3)
array([[152., 139., 131.],
         [154., 141., 133.],
         [156., 143., 135.],
         [157., 144., 136.],
         [166., 153., 145.],
         [168., 155., 147.],
         [171., 158., 150.],
         [174., 160., 149.],
         [175., 161., 150.],
         [176., 162., 151.],
         [177., 163., 152.],
         [173., 159., 148.],
         [173., 159., 148.],
         [173., 159., 148.],
         [178., 163., 156.],
         [177., 162., 155.],
         [175., 160., 153.],
         [170., 155., 148.],
         [168., 153., 146.],
         [170., 155., 148.],
         [173., 158., 151.],
         [174., 159., 152.],
         [177., 163., 154.],
         [179., 165., 156.],
         [174., 160., 151.],
         [172., 158., 149.],
         [177., 163., 154.],
         [177., 163., 154.],
         [174., 160., 151.],
         [173., 160., 151.],
         [172., 158., 145.],
         [166., 150., 134.],
         [149., 133., 117.],
         [ 72.,  54.,  40.],
         [ 40.,  22.,  12.],
         [ 33.,  13.,   6.],
         [ 33.,  16.,   8.],
         [ 35.,  18.,  10.],
         [ 35.,  18.,  10.],
         [ 34.,  17.,   9.],
         [ 34.,  17.,   9.],
         [ 35.,  18.,  10.],
         [ 37.,  20.,  12.],
         [ 35.,  16.,  12.],
         [ 34.,  15.,  11.],
         [ 33.,  14.,  10.],
         [ 33.,  14.,  10.],
         [ 34.,  15.,  11.],
         [ 34.,  15.,  11.],
         [ 31.,  12.,   8.],
         [ 33.,  14.,   7.],
         [ 33.,  14.,   7.],
         [ 33.,  14.,   7.],
         [ 33.,  14.,   7.],
         [ 34.,  15.,   8.],
         [ 34.,  15.,   8.],
         [ 34.,  15.,   8.],
         [ 34.,  15.,   8.],
         [ 34.,  15.,   8.],
         [ 35.,  16.,   9.],
         [ 36.,  17.,  10.],
         [ 35.,  16.,   9.],
         [ 35.,  16.,   9.],
         [ 35.,  16.,   9.],
         [ 36.,  17.,  10.],
         [ 34.,  15.,   8.],
         [ 33.,  14.,   7.],
         [ 32.,  13.,   6.],
         [ 29.,  10.,   3.],
         [ 33.,  14.,   7.],
         [ 33.,  14.,   7.],
         [ 32.,  13.,   6.],
         [ 33.,  14.,   7.],
         [ 33.,  14.,   7.],
         [ 33.,  14.,   7.],
         [ 34.,  15.,   8.],
         [ 31.,  12.,   5.],
         [ 30.,  11.,   4.],
         [ 34.,  15.,   8.],
         [ 31.,   7.,   3.],
         [ 33.,  12.,   9.],
         [ 32.,  14.,  10.],
         [ 26.,  12.,   9.],
         [ 17.,   7.,   5.],
         [ 12.,   4.,   1.],
         [ 13.,   8.,   4.],
         [ 21.,   7.,   4.],
         [ 19.,   5.,   2.],
         [ 21.,  10.,   6.],
         [ 27.,  14.,   8.],
         [ 31.,  12.,   6.],
         [ 33.,   8.,   4.],
         [ 36.,   6.,   4.],
         [ 30.,  11.,   5.],
         [ 31.,  10.,   5.],
         [ 33.,  12.,   7.],
         [ 35.,  11.,   7.],
         [ 33.,   9.,   5.],
         [ 34.,   9.,   5.],
         [ 34.,   9.,   5.],
         [ 37.,  10.,   3.],
         [ 37.,  10.,   3.],
         [ 38.,  11.,   4.],
         [ 39.,  12.,   5.],
         [ 37.,  10.,   3.],
         [ 37.,  10.,   3.],
         [ 37.,  10.,   3.],
         [ 37.,  10.,   3.],
         [ 36.,  14.,   3.],
         [ 35.,  13.,   2.],
         [ 34.,  12.,   1.],
         [ 34.,  12.,   1.],
         [ 33.,  11.,   0.],
         [ 36.,  14.,   3.],
         [ 39.,  17.,   6.],
         [ 41.,  14.,   5.],
         [ 40.,  13.,   4.],
         [ 39.,  12.,   3.],
         [ 39.,  12.,   3.],
         [ 40.,  13.,   4.],
         [ 40.,  13.,   4.],
         [ 37.,  10.,   1.],
         [ 38.,  11.,   2.],
         [ 39.,  12.,   3.],
         [ 40.,  13.,   4.],
         [ 39.,  12.,   3.],
         [ 39.,  12.,   3.],
         [ 38.,  11.,   2.],
         [ 37.,  10.,   1.],
         [ 40.,  11.,   5.],
         [ 37.,   8.,   2.],
         [ 37.,  10.,   3.],
         [ 40.,  13.,   6.],
         [ 36.,  11.,   4.],
         [ 34.,  11.,   3.],
         [ 33.,  10.,   2.],
         [ 33.,  12.,   7.],
         [ 33.,  12.,   7.],
         [ 31.,  10.,   5.],
         [ 30.,   9.,   4.],
         [ 31.,  10.,   5.],
         [ 32.,  11.,   6.],
         [ 33.,  12.,   7.],
         [ 32.,  12.,   3.],
         [ 32.,  12.,   3.],
         [ 32.,  12.,   3.],
         [ 32.,  12.,   3.],
         [ 31.,  11.,   2.],
         [ 30.,  10.,   1.],
         [ 29.,   9.,   0.],
         [ 29.,   9.,   0.],
         [ 30.,  10.,   3.],
         [ 31.,  11.,   4.],
         [ 32.,  12.,   5.],
         [ 29.,   9.,   2.],
         [ 30.,  10.,   3.],
         [ 28.,   8.,   1.],
         [ 27.,   7.,   0.],
         [ 28.,   9.,   2.],
         [ 27.,   8.,   1.],
         [ 26.,   7.,   0.],
         [ 30.,  11.,   4.],
         [ 30.,  11.,   4.],
         [ 30.,  11.,   4.],
         [ 30.,  11.,   4.],
         [ 30.,  11.,   7.],
         [ 28.,   9.,   5.],
         [ 27.,   8.,   4.],
         [ 27.,   8.,   4.],
         [ 27.,   8.,   4.],
         [ 27.,   8.,   4.],
         [ 27.,   8.,   4.],
         [ 27.,   9.,   7.],
         [ 27.,   9.,   7.],
         [ 27.,   9.,   7.],
         [ 27.,   9.,   7.],
         [ 25.,   7.,   5.],
         [ 25.,   7.,   5.],
         [ 25.,   7.,   5.],
         [ 23.,  10.,   1.],
         [ 23.,  10.,   1.],
         [ 26.,  12.,   3.],
         [ 26.,  12.,   3.],
         [ 29.,  12.,   4.],
         [ 30.,  11.,   4.],
         [ 30.,  11.,   4.],
         [ 31.,  11.,   4.],
         [ 31.,  11.,   4.],
         [ 30.,  10.,   3.],
         [ 29.,   9.,   2.],
         [ 30.,  10.,   3.],
         [ 30.,  10.,   3.],
         [ 30.,  10.,   3.],
         [ 30.,  10.,   3.],
         [ 24.,  10.,   7.],
         [ 23.,   9.,   6.],
         [ 22.,   8.,   5.],
         [ 22.,   8.,   5.],
         [ 24.,  10.,   7.],
         [ 23.,   9.,   6.],
         [ 22.,   8.,   5.],
         [ 19.,  10.,   5.],
         [ 19.,  10.,   5.],
         [ 19.,  10.,   5.],
         [ 19.,  10.,   5.],
         [ 19.,  10.,   5.],
         [ 19.,  10.,   5.],
         [ 19.,  10.,   5.],
         [ 18.,  10.,   7.],
         [ 17.,   9.,   6.],
         [ 16.,   8.,   5.],
         [ 17.,   9.,   6.],
         [ 17.,   9.,   6.],
         [ 16.,   8.,   5.],
         [ 15.,   7.,   4.],
         [ 13.,   9.,   8.],
         [  8.,   4.,   3.],
         [  9.,   5.,   4.],
         [ 13.,   9.,   8.],
         [ 10.,   6.,   5.],
         [ 13.,   9.,   8.],
         [ 14.,  10.,   9.],
         [  9.,   9.,   9.],
         [  9.,   9.,   9.]], dtype=float32)
In [46]:
from PIL import ImageFile                            
  ImageFile.LOAD_TRUNCATED_IMAGES = True                 
  
  # pre-process the data for Keras
  train_tensors = paths_to_tensor(train_files).astype('float32')/255
  valid_tensors = paths_to_tensor(valid_files).astype('float32')/255
  test_tensors = paths_to_tensor(test_files).astype('float32')/255
  
100%|██████████| 6680/6680 [01:36<00:00, 40.35it/s] 
  100%|██████████| 835/835 [00:11<00:00, 75.03it/s] 
  100%|██████████| 836/836 [00:10<00:00, 76.37it/s] 
  
In [47]:
# After process image data
  display(train_tensors[0][0].shape)
  display(train_tensors[0][0])
  
(224, 3)
array([[0.60784316, 0.61960787, 0.69411767],
         [0.6627451 , 0.6509804 , 0.7254902 ],
         [0.6392157 , 0.65882355, 0.7372549 ],
         [0.6117647 , 0.6313726 , 0.70980394],
         [0.63529414, 0.6392157 , 0.70980394],
         [0.63529414, 0.65882355, 0.7058824 ],
         [0.6156863 , 0.6392157 , 0.7019608 ],
         [0.60784316, 0.61960787, 0.6862745 ],
         [0.67058825, 0.6666667 , 0.7294118 ],
         [0.61960787, 0.64705884, 0.72156864],
         [0.61960787, 0.6392157 , 0.7176471 ],
         [0.61960787, 0.627451  , 0.7176471 ],
         [0.6       , 0.6117647 , 0.6862745 ],
         [0.6431373 , 0.654902  , 0.7137255 ],
         [0.6392157 , 0.64705884, 0.7294118 ],
         [0.62352943, 0.6509804 , 0.72156864],
         [0.654902  , 0.6666667 , 0.73333335],
         [0.6156863 , 0.63529414, 0.70980394],
         [0.6117647 , 0.6313726 , 0.7058824 ],
         [0.627451  , 0.64705884, 0.72156864],
         [0.63529414, 0.65882355, 0.72156864],
         [0.6431373 , 0.654902  , 0.72156864],
         [0.627451  , 0.6392157 , 0.69803923],
         [0.6313726 , 0.6431373 , 0.70980394],
         [0.6313726 , 0.6431373 , 0.70980394],
         [0.6392157 , 0.6313726 , 0.7176471 ],
         [0.61960787, 0.6392157 , 0.7137255 ],
         [0.62352943, 0.63529414, 0.7019608 ],
         [0.6509804 , 0.654902  , 0.7254902 ],
         [0.62352943, 0.6431373 , 0.72156864],
         [0.6392157 , 0.6509804 , 0.7254902 ],
         [0.6431373 , 0.654902  , 0.72156864],
         [0.6392157 , 0.6627451 , 0.7254902 ],
         [0.6313726 , 0.654902  , 0.7176471 ],
         [0.6392157 , 0.6627451 , 0.7254902 ],
         [0.6313726 , 0.654902  , 0.7176471 ],
         [0.6156863 , 0.6392157 , 0.69411767],
         [0.62352943, 0.6431373 , 0.7176471 ],
         [0.627451  , 0.64705884, 0.72156864],
         [0.6313726 , 0.654902  , 0.7176471 ],
         [0.62352943, 0.64705884, 0.70980394],
         [0.62352943, 0.6509804 , 0.7137255 ],
         [0.6509804 , 0.6627451 , 0.72156864],
         [0.64705884, 0.65882355, 0.7254902 ],
         [0.6392157 , 0.6509804 , 0.7176471 ],
         [0.63529414, 0.65882355, 0.7137255 ],
         [0.62352943, 0.64705884, 0.7019608 ],
         [0.63529414, 0.64705884, 0.7058824 ],
         [0.62352943, 0.6509804 , 0.72156864],
         [0.60784316, 0.6313726 , 0.69411767],
         [0.6117647 , 0.63529414, 0.69803923],
         [0.6156863 , 0.6392157 , 0.7019608 ],
         [0.62352943, 0.63529414, 0.7019608 ],
         [0.63529414, 0.6509804 , 0.69803923],
         [0.5568628 , 0.5803922 , 0.63529414],
         [0.5568628 , 0.58431375, 0.654902  ],
         [0.56078434, 0.57254905, 0.6392157 ],
         [0.54901963, 0.56078434, 0.627451  ],
         [0.5058824 , 0.5411765 , 0.6       ],
         [0.6039216 , 0.627451  , 0.6745098 ],
         [0.6156863 , 0.6431373 , 0.7058824 ],
         [0.63529414, 0.64705884, 0.7137255 ],
         [0.61960787, 0.6392157 , 0.7137255 ],
         [0.63529414, 0.654902  , 0.7294118 ],
         [0.6313726 , 0.654902  , 0.7019608 ],
         [0.6156863 , 0.6392157 , 0.69411767],
         [0.6156863 , 0.6392157 , 0.7019608 ],
         [0.6313726 , 0.6627451 , 0.7137255 ],
         [0.63529414, 0.654902  , 0.7294118 ],
         [0.63529414, 0.64705884, 0.7137255 ],
         [0.6392157 , 0.63529414, 0.7058824 ],
         [0.62352943, 0.6509804 , 0.7137255 ],
         [0.61960787, 0.64705884, 0.70980394],
         [0.627451  , 0.6509804 , 0.7137255 ],
         [0.61960787, 0.6431373 , 0.7058824 ],
         [0.63529414, 0.65882355, 0.72156864],
         [0.627451  , 0.6509804 , 0.7137255 ],
         [0.627451  , 0.6392157 , 0.7058824 ],
         [0.62352943, 0.64705884, 0.69411767],
         [0.61960787, 0.6392157 , 0.7137255 ],
         [0.61960787, 0.6431373 , 0.69803923],
         [0.627451  , 0.6509804 , 0.7137255 ],
         [0.6313726 , 0.654902  , 0.70980394],
         [0.61960787, 0.6431373 , 0.6901961 ],
         [0.6156863 , 0.63529414, 0.70980394],
         [0.61960787, 0.6431373 , 0.7058824 ],
         [0.62352943, 0.64705884, 0.7019608 ],
         [0.60784316, 0.6313726 , 0.69411767],
         [0.6156863 , 0.6431373 , 0.7058824 ],
         [0.5764706 , 0.6       , 0.6627451 ],
         [0.5803922 , 0.6039216 , 0.6509804 ],
         [0.54901963, 0.58431375, 0.6509804 ],
         [0.57254905, 0.6       , 0.6627451 ],
         [0.56078434, 0.58431375, 0.6392157 ],
         [0.5686275 , 0.5921569 , 0.64705884],
         [0.56078434, 0.57254905, 0.6313726 ],
         [0.5372549 , 0.5647059 , 0.627451  ],
         [0.5568628 , 0.5764706 , 0.6509804 ],
         [0.5137255 , 0.5372549 , 0.58431375],
         [0.3882353 , 0.40392157, 0.4509804 ],
         [0.4627451 , 0.45882353, 0.52156866],
         [0.4862745 , 0.49411765, 0.54509807],
         [0.54901963, 0.5686275 , 0.6431373 ],
         [0.5019608 , 0.5372549 , 0.6039216 ],
         [0.45490196, 0.48235294, 0.5529412 ],
         [0.49411765, 0.52156866, 0.59607846],
         [0.4392157 , 0.48235294, 0.56078434],
         [0.5058824 , 0.5254902 , 0.6117647 ],
         [0.5411765 , 0.5372549 , 0.6       ],
         [0.3529412 , 0.37254903, 0.39607844],
         [0.39607844, 0.42352942, 0.45490196],
         [0.4745098 , 0.5058824 , 0.5803922 ],
         [0.4862745 , 0.52156866, 0.5803922 ],
         [0.5254902 , 0.5529412 , 0.62352943],
         [0.5568628 , 0.5764706 , 0.6       ],
         [0.5176471 , 0.5254902 , 0.5764706 ],
         [0.20392157, 0.19215687, 0.23529412],
         [0.25490198, 0.27058825, 0.31764707],
         [0.3647059 , 0.37254903, 0.41960785],
         [0.00392157, 0.00392157, 0.01176471],
         [0.11372549, 0.08235294, 0.07058824],
         [0.4392157 , 0.44705883, 0.5058824 ],
         [0.4       , 0.42745098, 0.49019608],
         [0.14901961, 0.14117648, 0.19215687],
         [0.09803922, 0.10588235, 0.09411765],
         [0.23921569, 0.23921569, 0.27058825],
         [0.17254902, 0.16862746, 0.16078432],
         [0.22745098, 0.22745098, 0.1882353 ],
         [0.627451  , 0.6627451 , 0.7294118 ],
         [0.6039216 , 0.61960787, 0.6666667 ],
         [0.6313726 , 0.6431373 , 0.70980394],
         [0.6392157 , 0.654902  , 0.7019608 ],
         [0.62352943, 0.63529414, 0.7019608 ],
         [0.627451  , 0.6392157 , 0.7058824 ],
         [0.6039216 , 0.627451  , 0.6745098 ],
         [0.6039216 , 0.627451  , 0.68235296],
         [0.61960787, 0.6313726 , 0.69803923],
         [0.6392157 , 0.63529414, 0.7058824 ],
         [0.63529414, 0.6392157 , 0.70980394],
         [0.62352943, 0.627451  , 0.69803923],
         [0.6431373 , 0.6392157 , 0.7019608 ],
         [0.6392157 , 0.6431373 , 0.72156864],
         [0.6431373 , 0.627451  , 0.7176471 ],
         [0.62352943, 0.654902  , 0.69803923],
         [0.63529414, 0.64705884, 0.7058824 ],
         [0.63529414, 0.6392157 , 0.70980394],
         [0.61960787, 0.6313726 , 0.6901961 ],
         [0.60784316, 0.6313726 , 0.6862745 ],
         [0.60784316, 0.63529414, 0.7058824 ],
         [0.6039216 , 0.6156863 , 0.6901961 ],
         [0.62352943, 0.627451  , 0.69803923],
         [0.6156863 , 0.627451  , 0.7019608 ],
         [0.6156863 , 0.6392157 , 0.6862745 ],
         [0.6117647 , 0.6392157 , 0.7019608 ],
         [0.6117647 , 0.63529414, 0.69803923],
         [0.6117647 , 0.6392157 , 0.70980394],
         [0.62352943, 0.6313726 , 0.7137255 ],
         [0.61960787, 0.6313726 , 0.69803923],
         [0.62352943, 0.627451  , 0.69803923],
         [0.6039216 , 0.6313726 , 0.69411767],
         [0.6039216 , 0.63529414, 0.6862745 ],
         [0.6156863 , 0.6392157 , 0.69411767],
         [0.60784316, 0.6313726 , 0.6862745 ],
         [0.6039216 , 0.6313726 , 0.69411767],
         [0.6       , 0.62352943, 0.6862745 ],
         [0.6       , 0.62352943, 0.6862745 ],
         [0.60784316, 0.6313726 , 0.69411767],
         [0.6117647 , 0.63529414, 0.69803923],
         [0.6156863 , 0.627451  , 0.7019608 ],
         [0.6039216 , 0.6156863 , 0.6901961 ],
         [0.6156863 , 0.62352943, 0.7058824 ],
         [0.61960787, 0.62352943, 0.69411767],
         [0.6117647 , 0.62352943, 0.6901961 ],
         [0.6       , 0.62352943, 0.6862745 ],
         [0.59607846, 0.61960787, 0.68235296],
         [0.59607846, 0.62352943, 0.6862745 ],
         [0.58431375, 0.6117647 , 0.6745098 ],
         [0.60784316, 0.6117647 , 0.68235296],
         [0.6117647 , 0.6156863 , 0.6862745 ],
         [0.6039216 , 0.6156863 , 0.68235296],
         [0.6       , 0.6117647 , 0.6784314 ],
         [0.5882353 , 0.6117647 , 0.6745098 ],
         [0.5921569 , 0.6156863 , 0.6784314 ],
         [0.6       , 0.6117647 , 0.6784314 ],
         [0.6       , 0.6117647 , 0.67058825],
         [0.6       , 0.6117647 , 0.67058825],
         [0.5921569 , 0.61960787, 0.68235296],
         [0.5882353 , 0.6156863 , 0.6862745 ],
         [0.59607846, 0.60784316, 0.6666667 ],
         [0.59607846, 0.60784316, 0.6666667 ],
         [0.5803922 , 0.6039216 , 0.6666667 ],
         [0.6       , 0.6039216 , 0.6745098 ],
         [0.59607846, 0.60784316, 0.6745098 ],
         [0.58431375, 0.60784316, 0.67058825],
         [0.5882353 , 0.6117647 , 0.6745098 ],
         [0.58431375, 0.6117647 , 0.6745098 ],
         [0.58431375, 0.60784316, 0.67058825],
         [0.5921569 , 0.6156863 , 0.6784314 ],
         [0.58431375, 0.60784316, 0.67058825],
         [0.5764706 , 0.6       , 0.6627451 ],
         [0.59607846, 0.60784316, 0.6745098 ],
         [0.59607846, 0.6       , 0.67058825],
         [0.5921569 , 0.6039216 , 0.67058825],
         [0.5803922 , 0.60784316, 0.67058825],
         [0.5921569 , 0.60784316, 0.654902  ],
         [0.5882353 , 0.6       , 0.6666667 ],
         [0.5882353 , 0.6       , 0.6745098 ],
         [0.58431375, 0.6117647 , 0.68235296],
         [0.5647059 , 0.6       , 0.6666667 ],
         [0.5882353 , 0.6       , 0.6745098 ],
         [0.59607846, 0.6       , 0.67058825],
         [0.5764706 , 0.5882353 , 0.654902  ],
         [0.57254905, 0.6       , 0.6627451 ],
         [0.5803922 , 0.60784316, 0.68235296],
         [0.5764706 , 0.6039216 , 0.6784314 ],
         [0.5686275 , 0.59607846, 0.6666667 ],
         [0.56078434, 0.5882353 , 0.65882355],
         [0.5803922 , 0.6       , 0.6745098 ],
         [0.58431375, 0.6039216 , 0.6901961 ],
         [0.57254905, 0.6       , 0.6627451 ],
         [0.5803922 , 0.6156863 , 0.6745098 ],
         [0.5882353 , 0.6       , 0.6745098 ],
         [0.57254905, 0.58431375, 0.65882355],
         [0.5529412 , 0.5803922 , 0.6509804 ]], dtype=float32)

(IMPLEMENTATION) Model Architecture¶

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    model.summary()
  
  

We have imported some Python modules to get you started, but feel free to import as many modules as you need. If you end up getting stuck, here's a hint that specifies a model that trains relatively fast on CPU and attains >1% test accuracy in 5 epochs:

Sample CNN

Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. If you chose to use the hinted architecture above, describe why you think that CNN architecture should work well for the image classification task.

Answer:

Here I choose to set kernel_initialier to he_normal but not the default glorot_uniform. Though glorot_uniform , the Xavier Glorot uniform initialization method, which is perfectly fine for the majority of tasks; however, for deeper neural networks you may want to use he_normal (MSRA/He et al. initialization) which works especially well when our network has a large number of parameters (i.e., VGGNet).

ReLU activation is applied along with batch normalization and dropout. I choose to use batch normalization because it tends to stabilize training and make tuning hyperparameters easier.

Dropout’s purpose is to help our network generalize and not overfit. Neurons from the current layer, with probability p, will randomly disconnect from neurons in the next layer so that the network has to rely on the existing connections.

Finally, I tend to decide this model relatively small becuase I want to train faster and leave some time to do the Augmention part.

In [79]:
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
  from keras.layers import Dropout, Flatten, Dense
  from keras.models import Sequential
  
  from keras.layers.normalization import BatchNormalization
  
  model = Sequential()
  
  ### TODO: Define your architecture.
  
   
  # Layer 1
  model.add(
      Conv2D(
          filters=64, 
          kernel_size=2,
          kernel_initializer="he_normal",
          padding='same', 
          activation='elu', 
          input_shape=(224, 224, 3))
  )
  model.add(BatchNormalization())
  model.add(MaxPooling2D(pool_size=4))
  model.add(Dropout(0.25))
  
  # Layer 2
  model.add(Conv2D(
      filters=128, 
      kernel_size=2, 
      kernel_initializer="he_normal",
      padding='same', 
      activation='elu'
  ))
  model.add(BatchNormalization())
  model.add(MaxPooling2D(pool_size=4))
  model.add(Dropout(0.25))
  
  
  # Flatten will result in a larger Dense layer afterwards, which is more expensive and may result in worse overfitting. 
  # But if you have lots of data, it might also perform better.
  # model.add(Flatten())
  
  # fully-connected layer
  model.add(GlobalAveragePooling2D())
  
  model.add(Dense(512, activation='relu'))
  model.add(BatchNormalization())
  model.add(Dropout(0.5))
  
  # This is because we have 133 dogs labels
  model.add(Dense(133, activation='softmax'))
  model.summary()
  
_________________________________________________________________
  Layer (type)                 Output Shape              Param #   
  =================================================================
  conv2d_31 (Conv2D)           (None, 224, 224, 64)      832       
  _________________________________________________________________
  batch_normalization_43 (Batc (None, 224, 224, 64)      256       
  _________________________________________________________________
  max_pooling2d_28 (MaxPooling (None, 56, 56, 64)        0         
  _________________________________________________________________
  dropout_43 (Dropout)         (None, 56, 56, 64)        0         
  _________________________________________________________________
  conv2d_32 (Conv2D)           (None, 56, 56, 128)       32896     
  _________________________________________________________________
  batch_normalization_44 (Batc (None, 56, 56, 128)       512       
  _________________________________________________________________
  max_pooling2d_29 (MaxPooling (None, 14, 14, 128)       0         
  _________________________________________________________________
  dropout_44 (Dropout)         (None, 14, 14, 128)       0         
  _________________________________________________________________
  global_average_pooling2d_13  (None, 128)               0         
  _________________________________________________________________
  dense_25 (Dense)             (None, 512)               66048     
  _________________________________________________________________
  batch_normalization_45 (Batc (None, 512)               2048      
  _________________________________________________________________
  dropout_45 (Dropout)         (None, 512)               0         
  _________________________________________________________________
  dense_26 (Dense)             (None, 133)               68229     
  =================================================================
  Total params: 170,821
  Trainable params: 169,413
  Non-trainable params: 1,408
  _________________________________________________________________
  

The difference between Flatten() and GlobalAveragePooling2D() in keras¶

That both seem to work doesn't mean they do the same.

Flatten will take a tensor of any shape and transform it into a one dimensional tensor (plus the samples dimension) but keeping all values in the tensor. For example a tensor (samples, 10, 20, 1) will be flattened to (samples, 10 20 1).

GlobalAveragePooling2D does something different. It applies average pooling on the spatial dimensions until each spatial dimension is one, and leaves other dimensions unchanged. In this case values are not kept as they are averaged. For example a tensor (samples, 10, 20, 1) would be output as (samples, 1, 1, 1), assuming the 2nd and 3rd dimensions were spatial (channels last).

Compile the Model¶

In [80]:
model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
  

(IMPLEMENTATION) Train the Model¶

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

[Bonus] Augmented Images¶

Sai: I chose to use augmentation here!

In [81]:
from keras.preprocessing.image import ImageDataGenerator
  
  # create and configure augmented image generator
  datagen_train = ImageDataGenerator(
      width_shift_range=0.2,  # randomly shift images horizontally (20% of total width)
      height_shift_range=0.2,  # randomly shift images vertically (20% of total height)
      horizontal_flip=True) # randomly flip images horizontally
  
  # create and configure augmented image generator
  datagen_valid = ImageDataGenerator(
      width_shift_range=0.2,  # randomly shift images horizontally (20% of total width)
      height_shift_range=0.2,  # randomly shift images vertically (20% of total height)
      horizontal_flip=True) # randomly flip images horizontally
  
  # fit augmented image generator on data
  datagen_train.fit(train_tensors)
  datagen_valid.fit(valid_tensors)
  

[Bonus] Visualize Original and Augmented Images¶

In [82]:
# take subset of training data
  train_tensors_subset = train_tensors[:12]
  
  # visualize subset of training data
  fig = plt.figure(figsize=(20,2))
  for i in range(0, len(train_tensors_subset)):
      ax = fig.add_subplot(1, 12, i+1)
      ax.imshow(train_tensors_subset[i])
  fig.suptitle('Subset of Original Training Images', fontsize=20)
  plt.show()
  
  # visualize augmented images
  fig = plt.figure(figsize=(20,2))
  for x_batch in datagen_train.flow(train_tensors_subset, batch_size=12):
      for i in range(0, 12):
          ax = fig.add_subplot(1, 12, i+1)
          ax.imshow(x_batch[i])
      fig.suptitle('Augmented Images', fontsize=20)
      plt.show()
      break;
  
In [83]:
from keras.callbacks import ModelCheckpoint  
  from keras_tqdm import TQDMNotebookCallback
  
  ### TODO: specify the number of epochs that you would like to use to train the model.
  epochs = 20
  batch_size = 4
  
  ### Do NOT modify the code below this line.
  checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch.hdf5', 
                                 verbose=1, save_best_only=True)
  
  from_scratch_model_history = model.fit_generator(datagen_train.flow(train_tensors, train_targets, batch_size=batch_size),
                      steps_per_epoch=train_tensors.shape[0] // batch_size,
                      epochs=epochs, verbose=2, callbacks=[checkpointer, TQDMNotebookCallback()],
                      validation_data=datagen_valid.flow(valid_tensors, valid_targets, batch_size=batch_size),
                      validation_steps=train_tensors.shape[0] // batch_size)
  
  '''
  Original code that I didn't use "Data Augmentation"
  model.fit(
      train_tensors, 
      train_targets, 
      validation_data=(valid_tensors, valid_targets),
      epochs=epochs, 
      batch_size=20, 
      callbacks=[checkpointer], 
      verbose=1
  )
  '''
  
Epoch 1/20
  
 - 116s - loss: 5.1401 - acc: 0.0097 - val_loss: 5.4367 - val_acc: 0.0087
  
  Epoch 00001: val_loss improved from inf to 5.43667, saving model to saved_models/weights.best.from_scratch.hdf5
  Epoch 2/20
  
 - 118s - loss: 4.9273 - acc: 0.0096 - val_loss: 5.6205 - val_acc: 0.0124
  
  Epoch 00002: val_loss did not improve from 5.43667
  Epoch 3/20
  
 - 120s - loss: 4.8920 - acc: 0.0115 - val_loss: 5.2233 - val_acc: 0.0096
  
  Epoch 00003: val_loss improved from 5.43667 to 5.22332, saving model to saved_models/weights.best.from_scratch.hdf5
  Epoch 4/20
  
 - 120s - loss: 4.8818 - acc: 0.0172 - val_loss: 5.0224 - val_acc: 0.0100
  
  Epoch 00004: val_loss improved from 5.22332 to 5.02236, saving model to saved_models/weights.best.from_scratch.hdf5
  Epoch 5/20
  
 - 122s - loss: 4.8936 - acc: 0.0132 - val_loss: 5.0062 - val_acc: 0.0177
  
  Epoch 00005: val_loss improved from 5.02236 to 5.00624, saving model to saved_models/weights.best.from_scratch.hdf5
  Epoch 6/20
  
 - 123s - loss: 4.8896 - acc: 0.0129 - val_loss: 5.1688 - val_acc: 0.0168
  
  Epoch 00006: val_loss did not improve from 5.00624
  Epoch 7/20
  
 - 126s - loss: 4.9000 - acc: 0.0136 - val_loss: 5.1704 - val_acc: 0.0153
  
  Epoch 00007: val_loss did not improve from 5.00624
  Epoch 8/20
  
 - 124s - loss: 4.9024 - acc: 0.0148 - val_loss: 5.1374 - val_acc: 0.0111
  
  Epoch 00008: val_loss did not improve from 5.00624
  Epoch 9/20
  
 - 121s - loss: 4.9087 - acc: 0.0139 - val_loss: 5.1013 - val_acc: 0.0115
  
  Epoch 00009: val_loss did not improve from 5.00624
  Epoch 10/20
  
 - 124s - loss: 4.9145 - acc: 0.0141 - val_loss: 5.0994 - val_acc: 0.0192
  
  Epoch 00010: val_loss did not improve from 5.00624
  Epoch 11/20
  
 - 124s - loss: 4.9084 - acc: 0.0150 - val_loss: 5.1220 - val_acc: 0.0163
  
  Epoch 00011: val_loss did not improve from 5.00624
  Epoch 12/20
  
 - 126s - loss: 4.9126 - acc: 0.0157 - val_loss: 5.1580 - val_acc: 0.0213
  
  Epoch 00012: val_loss did not improve from 5.00624
  Epoch 13/20
  
 - 128s - loss: 4.9174 - acc: 0.0180 - val_loss: 5.2258 - val_acc: 0.0171
  
  Epoch 00013: val_loss did not improve from 5.00624
  Epoch 14/20
  
 - 125s - loss: 4.9220 - acc: 0.0145 - val_loss: 5.1309 - val_acc: 0.0168
  
  Epoch 00014: val_loss did not improve from 5.00624
  Epoch 15/20
  
 - 129s - loss: 4.9139 - acc: 0.0153 - val_loss: 5.3444 - val_acc: 0.0181
  
  Epoch 00015: val_loss did not improve from 5.00624
  Epoch 16/20
  
 - 129s - loss: 4.9190 - acc: 0.0141 - val_loss: 5.4444 - val_acc: 0.0106
  
  Epoch 00016: val_loss did not improve from 5.00624
  Epoch 17/20
  
 - 127s - loss: 4.9215 - acc: 0.0157 - val_loss: 5.1924 - val_acc: 0.0193
  
  Epoch 00017: val_loss did not improve from 5.00624
  Epoch 18/20
  
 - 131s - loss: 4.9244 - acc: 0.0154 - val_loss: 5.0645 - val_acc: 0.0175
  
  Epoch 00018: val_loss did not improve from 5.00624
  Epoch 19/20
  
 - 133s - loss: 4.9257 - acc: 0.0151 - val_loss: 5.1664 - val_acc: 0.0148
  
  Epoch 00019: val_loss did not improve from 5.00624
  Epoch 20/20
  
 - 130s - loss: 4.9252 - acc: 0.0151 - val_loss: 5.0137 - val_acc: 0.0247
  
  Epoch 00020: val_loss did not improve from 5.00624
  
Out[83]:
'\nOriginal code that I didn\'t use "Data Augmentation"\nmodel.fit(\n    train_tensors, \n    train_targets, \n    validation_data=(valid_tensors, valid_targets),\n    epochs=epochs, \n    batch_size=20, \n    callbacks=[checkpointer], \n    verbose=1\n)\n'

[Bonus] Visualize performances¶

Sai: SInce I want to check the performance of my model, I write this helper class to check.

In [84]:
def show_train_history(train_history, train, validation):
      plt.plot(train_history.history[train])
      plt.plot(train_history.history[validation])
      plt.title('Train History')
      plt.ylabel('train')
      plt.xlabel('Epoch')
      plt.legend(['train', 'validation'], loc='center right')
      plt.show()
  
  show_train_history(from_scratch_model_history, 'acc', 'val_acc')
  show_train_history(from_scratch_model_history, 'loss', 'val_loss')
  

Load the Model with the Best Validation Loss¶

In [85]:
model.load_weights('saved_models/weights.best.from_scratch.hdf5')
  

Test the Model¶

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 1%.

In [86]:
# get index of predicted dog breed for each image in test set
  dog_breed_predictions = [np.argmax(model.predict(np.expand_dims(tensor, axis=0))) for tensor in test_tensors]
  
  # report test accuracy
  test_accuracy = 100*np.sum(np.array(dog_breed_predictions)==np.argmax(test_targets, axis=1))/len(dog_breed_predictions)
  print('Test accuracy: %.4f%%' % test_accuracy)
  
Test accuracy: 1.6746%
  

Step 4: Use a CNN to Classify Dog Breeds¶

To reduce training time without sacrificing accuracy, we show you how to train a CNN using transfer learning. In the following step, you will get a chance to use transfer learning to train your own CNN.

Obtain Bottleneck Features¶

In [26]:
bottleneck_features = np.load('bottleneck_features/DogVGG16Data.npz')
  train_VGG16 = bottleneck_features['train']
  valid_VGG16 = bottleneck_features['valid']
  test_VGG16 = bottleneck_features['test']
  

Model Architecture¶

The model uses the the pre-trained VGG-16 model as a fixed feature extractor, where the last convolutional output of VGG-16 is fed as input to our model. We only add a global average pooling layer and a fully connected layer, where the latter contains one node for each dog category and is equipped with a softmax.

In [27]:
VGG16_model = Sequential()
  VGG16_model.add(GlobalAveragePooling2D(input_shape=train_VGG16.shape[1:]))
  VGG16_model.add(Dense(133, activation='softmax'))
  
  VGG16_model.summary()
  
_________________________________________________________________
  Layer (type)                 Output Shape              Param #   
  =================================================================
  global_average_pooling2d_4 ( (None, 512)               0         
  _________________________________________________________________
  dense_10 (Dense)             (None, 133)               68229     
  =================================================================
  Total params: 68,229
  Trainable params: 68,229
  Non-trainable params: 0
  _________________________________________________________________
  

Compile the Model¶

In [28]:
VGG16_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
  

Train the Model¶

In [29]:
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.VGG16.hdf5', 
                                 verbose=1, save_best_only=True)
  
  VGG16_model.fit(train_VGG16, train_targets, 
            validation_data=(valid_VGG16, valid_targets),
            epochs=20, batch_size=20, callbacks=[checkpointer], verbose=1)
  
Train on 6680 samples, validate on 835 samples
  Epoch 1/20
  6680/6680 [==============================] - 2s 244us/step - loss: 12.0936 - acc: 0.1216 - val_loss: 10.7365 - val_acc: 0.1976
  
  Epoch 00001: val_loss improved from inf to 10.73653, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 2/20
  6680/6680 [==============================] - 1s 160us/step - loss: 9.8120 - acc: 0.2844 - val_loss: 9.7915 - val_acc: 0.2790
  
  Epoch 00002: val_loss improved from 10.73653 to 9.79154, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 3/20
  6680/6680 [==============================] - 1s 157us/step - loss: 9.1214 - acc: 0.3548 - val_loss: 9.3883 - val_acc: 0.3174
  
  Epoch 00003: val_loss improved from 9.79154 to 9.38829, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 4/20
  6680/6680 [==============================] - 1s 162us/step - loss: 8.7054 - acc: 0.3960 - val_loss: 9.0642 - val_acc: 0.3365
  
  Epoch 00004: val_loss improved from 9.38829 to 9.06422, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 5/20
  6680/6680 [==============================] - 1s 167us/step - loss: 8.3162 - acc: 0.4269 - val_loss: 8.8655 - val_acc: 0.3341
  
  Epoch 00005: val_loss improved from 9.06422 to 8.86549, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 6/20
  6680/6680 [==============================] - 1s 155us/step - loss: 7.9520 - acc: 0.4567 - val_loss: 8.6572 - val_acc: 0.3509
  
  Epoch 00006: val_loss improved from 8.86549 to 8.65718, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 7/20
  6680/6680 [==============================] - 1s 152us/step - loss: 7.7211 - acc: 0.4781 - val_loss: 8.4018 - val_acc: 0.3880
  
  Epoch 00007: val_loss improved from 8.65718 to 8.40177, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 8/20
  6680/6680 [==============================] - 1s 154us/step - loss: 7.5608 - acc: 0.4993 - val_loss: 8.3236 - val_acc: 0.3916
  
  Epoch 00008: val_loss improved from 8.40177 to 8.32362, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 9/20
  6680/6680 [==============================] - 1s 154us/step - loss: 7.3817 - acc: 0.5099 - val_loss: 8.0817 - val_acc: 0.4108
  
  Epoch 00009: val_loss improved from 8.32362 to 8.08167, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 10/20
  6680/6680 [==============================] - 1s 186us/step - loss: 7.1997 - acc: 0.5320 - val_loss: 8.0929 - val_acc: 0.3976
  
  Epoch 00010: val_loss did not improve from 8.08167
  Epoch 11/20
  6680/6680 [==============================] - 1s 172us/step - loss: 7.1549 - acc: 0.5341 - val_loss: 8.0503 - val_acc: 0.4120
  
  Epoch 00011: val_loss improved from 8.08167 to 8.05030, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 12/20
  6680/6680 [==============================] - 1s 176us/step - loss: 7.0633 - acc: 0.5439 - val_loss: 8.0615 - val_acc: 0.4108
  
  Epoch 00012: val_loss did not improve from 8.05030
  Epoch 13/20
  6680/6680 [==============================] - 1s 165us/step - loss: 6.8749 - acc: 0.5549 - val_loss: 7.8112 - val_acc: 0.4228
  
  Epoch 00013: val_loss improved from 8.05030 to 7.81117, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 14/20
  6680/6680 [==============================] - 1s 153us/step - loss: 6.7577 - acc: 0.5662 - val_loss: 7.7422 - val_acc: 0.4383
  
  Epoch 00014: val_loss improved from 7.81117 to 7.74220, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 15/20
  6680/6680 [==============================] - 1s 180us/step - loss: 6.6402 - acc: 0.5722 - val_loss: 7.8006 - val_acc: 0.4228
  
  Epoch 00015: val_loss did not improve from 7.74220
  Epoch 16/20
  6680/6680 [==============================] - 1s 164us/step - loss: 6.3602 - acc: 0.5871 - val_loss: 7.4326 - val_acc: 0.4515
  
  Epoch 00016: val_loss improved from 7.74220 to 7.43262, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 17/20
  6680/6680 [==============================] - 1s 173us/step - loss: 6.1627 - acc: 0.5996 - val_loss: 7.2480 - val_acc: 0.4611
  
  Epoch 00017: val_loss improved from 7.43262 to 7.24805, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 18/20
  6680/6680 [==============================] - 1s 172us/step - loss: 5.9852 - acc: 0.6114 - val_loss: 7.1342 - val_acc: 0.4623
  
  Epoch 00018: val_loss improved from 7.24805 to 7.13422, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 19/20
  6680/6680 [==============================] - 1s 173us/step - loss: 5.8194 - acc: 0.6226 - val_loss: 7.1029 - val_acc: 0.4647
  
  Epoch 00019: val_loss improved from 7.13422 to 7.10287, saving model to saved_models/weights.best.VGG16.hdf5
  Epoch 20/20
  6680/6680 [==============================] - 1s 181us/step - loss: 5.7171 - acc: 0.6313 - val_loss: 6.8840 - val_acc: 0.4862
  
  Epoch 00020: val_loss improved from 7.10287 to 6.88396, saving model to saved_models/weights.best.VGG16.hdf5
  
Out[29]:
<keras.callbacks.History at 0x7f9b3ae83be0>

Load the Model with the Best Validation Loss¶

In [31]:
VGG16_model.load_weights('saved_models/weights.best.VGG16.hdf5')
  

Test the Model¶

Now, we can use the CNN to test how well it identifies breed within our test dataset of dog images. We print the test accuracy below.

In [32]:
# get index of predicted dog breed for each image in test set
  VGG16_predictions = [np.argmax(VGG16_model.predict(np.expand_dims(feature, axis=0))) for feature in test_VGG16]
  
  # report test accuracy
  test_accuracy = 100*np.sum(np.array(VGG16_predictions)==np.argmax(test_targets, axis=1))/len(VGG16_predictions)
  print('Test accuracy: %.4f%%' % test_accuracy)
  
Test accuracy: 49.1627%
  

Predict Dog Breed with the Model¶

In [33]:
from extract_bottleneck_features import *
  
  def VGG16_predict_breed(img_path):
      # extract bottleneck features
      bottleneck_feature = extract_VGG16(path_to_tensor(img_path))
      # obtain predicted vector
      predicted_vector = VGG16_model.predict(bottleneck_feature)
      # return dog breed that is predicted by the model
      return dog_names[np.argmax(predicted_vector)]
  

Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)¶

You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.

In Step 4, we used transfer learning to create a CNN using VGG-16 bottleneck features. In this section, you must use the bottleneck features from a different pre-trained model. To make things easier for you, we have pre-computed the features for all of the networks that are currently available in Keras:

The files are encoded as such:

Dog{network}Data.npz
  
  

where {network}, in the above filename, can be one of VGG19, Resnet50, InceptionV3, or Xception. Pick one of the above architectures, download the corresponding bottleneck features, and store the downloaded file in the bottleneck_features/ folder in the repository.

(IMPLEMENTATION) Obtain Bottleneck Features¶

In the code block below, extract the bottleneck features corresponding to the train, test, and validation sets by running the following:

bottleneck_features = np.load('bottleneck_features/Dog{network}Data.npz')
  train_{network} = bottleneck_features['train']
  valid_{network} = bottleneck_features['valid']
  test_{network} = bottleneck_features['test']
In [87]:
### TODO: Obtain bottleneck features from another pre-trained CNN.
  bottleneck_features = np.load('bottleneck_features/DogInceptionV3Data.npz')
  train_inceptionV3  = bottleneck_features['train']
  valid_inceptionV3  = bottleneck_features['valid']
  test_inceptionV3  = bottleneck_features['test']
  

(IMPLEMENTATION) Model Architecture¶

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    <your model's name>.summary()
  
  

Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.

Answer:

Why InceptionV3?¶

I have 2 main reasons to select InceptionV3 as my pre-trained structure

  • The number of params of InceptionV3 is the smallest among the others, which means it can train faster.
  • The Top-1 accurancy (only needs to predict once to get the correct answer) is the highest among those 4 options different_struture ref: Neural Network Architectures Keras Applications_tables

Why Batch Normalization after each ReLU activation?¶

I use batch normalization because it tends to stabilize training and make tuning hyperparameters easier. That said, it can double or triple your training time, but I think I can use it in this case since we only have the last 3 fully connected layers to train!

Also, the general use case is to use Batch Normalization between the linear and non-linear layers in your network, because it normalizes the input to your activation function, so that you're centered in the linear section of the activation function.

Why Dropout in each layers?¶

Dropout’s purpose is to help my network generalize and not overfit. Neurons from the current layer, with probability p, will randomly disconnect from neurons in the next layer so that the network has to rely on the existing connections. I highly recommend utilizing dropout.

In [89]:
### TODO: Define your architecture.
  inceptionV3_model = Sequential()
  inceptionV3_model.add(GlobalAveragePooling2D(input_shape=(5, 5, 2048)))
  
  inceptionV3_model.add(Dense(1024, activation='relu', kernel_initializer="he_normal"))
  inceptionV3_model.add(BatchNormalization())
  inceptionV3_model.add(Dropout(0.6))
  
  inceptionV3_model.add(Dense(512, activation='relu', kernel_initializer="he_normal"))
  inceptionV3_model.add(BatchNormalization())
  inceptionV3_model.add(Dropout(0.6))
  
  inceptionV3_model.add(Dense(256, activation='relu', kernel_initializer="he_normal"))
  inceptionV3_model.add(BatchNormalization())
  inceptionV3_model.add(Dropout(0.4))
  
  inceptionV3_model.add(Dense(128, activation='relu', kernel_initializer="he_normal"))
  inceptionV3_model.add(BatchNormalization())
  inceptionV3_model.add(Dropout(0.4))
  
  # 133 different labels
  inceptionV3_model.add(Dense(133, activation='softmax'))
  inceptionV3_model.summary()
  
_________________________________________________________________
  Layer (type)                 Output Shape              Param #   
  =================================================================
  global_average_pooling2d_15  (None, 2048)              0         
  _________________________________________________________________
  dense_32 (Dense)             (None, 1024)              2098176   
  _________________________________________________________________
  batch_normalization_50 (Batc (None, 1024)              4096      
  _________________________________________________________________
  dropout_50 (Dropout)         (None, 1024)              0         
  _________________________________________________________________
  dense_33 (Dense)             (None, 512)               524800    
  _________________________________________________________________
  batch_normalization_51 (Batc (None, 512)               2048      
  _________________________________________________________________
  dropout_51 (Dropout)         (None, 512)               0         
  _________________________________________________________________
  dense_34 (Dense)             (None, 256)               131328    
  _________________________________________________________________
  batch_normalization_52 (Batc (None, 256)               1024      
  _________________________________________________________________
  dropout_52 (Dropout)         (None, 256)               0         
  _________________________________________________________________
  dense_35 (Dense)             (None, 128)               32896     
  _________________________________________________________________
  batch_normalization_53 (Batc (None, 128)               512       
  _________________________________________________________________
  dropout_53 (Dropout)         (None, 128)               0         
  _________________________________________________________________
  dense_36 (Dense)             (None, 133)               17157     
  =================================================================
  Total params: 2,812,037
  Trainable params: 2,808,197
  Non-trainable params: 3,840
  _________________________________________________________________
  

(IMPLEMENTATION) Compile the Model¶

In [90]:
from keras import optimizers
  
  ### TODO: Compile the model.
  sgd = optimizers.SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
  inceptionV3_model.compile(loss='categorical_crossentropy', optimizer=sgd, 
                    metrics=['accuracy'])
  

(IMPLEMENTATION) Train the Model¶

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [92]:
### TODO: Train the model.
  #epochs = 20
  #batch_size = 8
  
  checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.InceptionV3.hdf5', verbose=1, 
                                 save_best_only=True)
  
  """
  "Data Augmentation"
  inceptionV3_model_train_history = inceptionV3_model.fit_generator(
      datagen_train.flow(train_inceptionV3, train_targets, batch_size=batch_size),
                      steps_per_epoch=train_inceptionV3.shape[0] // batch_size,
                      epochs=epochs, verbose=2, callbacks=[checkpointer],
                      validation_data=datagen_valid.flow(valid_inceptionV3, valid_targets, batch_size=batch_size),
                      validation_steps=train_inceptionV3.shape[0] // batch_size)
  
  """
  inceptionV3_model_train_history = inceptionV3_model.fit(
      train_inceptionV3, 
      train_targets, 
      epochs=20, 
      validation_data=(valid_inceptionV3, valid_targets), 
      callbacks=[checkpointer, TQDMNotebookCallback()], verbose=1, shuffle=True
  )
  
Train on 6680 samples, validate on 835 samples
  
Epoch 1/20
  
6680/6680 [==============================] - 3s 412us/step - loss: 1.7046 - acc: 0.5394 - val_loss: 0.7874 - val_acc: 0.7593
  
  Epoch 00001: val_loss improved from inf to 0.78740, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 2/20
  
6680/6680 [==============================] - 3s 406us/step - loss: 1.2831 - acc: 0.6283 - val_loss: 0.6887 - val_acc: 0.7737
  
  Epoch 00002: val_loss improved from 0.78740 to 0.68870, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 3/20
  
6680/6680 [==============================] - 3s 409us/step - loss: 1.0906 - acc: 0.6751 - val_loss: 0.6304 - val_acc: 0.7964
  
  Epoch 00003: val_loss improved from 0.68870 to 0.63041, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 4/20
  
6680/6680 [==============================] - 3s 405us/step - loss: 0.9910 - acc: 0.6987 - val_loss: 0.6124 - val_acc: 0.7988
  
  Epoch 00004: val_loss improved from 0.63041 to 0.61240, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 5/20
  
6680/6680 [==============================] - 3s 401us/step - loss: 0.9113 - acc: 0.7196 - val_loss: 0.5878 - val_acc: 0.8072
  
  Epoch 00005: val_loss improved from 0.61240 to 0.58780, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 6/20
  
6680/6680 [==============================] - 3s 403us/step - loss: 0.8764 - acc: 0.7320 - val_loss: 0.5830 - val_acc: 0.8108
  
  Epoch 00006: val_loss improved from 0.58780 to 0.58300, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 7/20
  
6680/6680 [==============================] - 3s 402us/step - loss: 0.8130 - acc: 0.7437 - val_loss: 0.5746 - val_acc: 0.8168
  
  Epoch 00007: val_loss improved from 0.58300 to 0.57456, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 8/20
  
6680/6680 [==============================] - 3s 397us/step - loss: 0.7761 - acc: 0.7560 - val_loss: 0.5741 - val_acc: 0.8168
  
  Epoch 00008: val_loss improved from 0.57456 to 0.57409, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 9/20
  
6680/6680 [==============================] - 3s 399us/step - loss: 0.7546 - acc: 0.7609 - val_loss: 0.5663 - val_acc: 0.8240
  
  Epoch 00009: val_loss improved from 0.57409 to 0.56626, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 10/20
  
6680/6680 [==============================] - 3s 400us/step - loss: 0.7177 - acc: 0.7692 - val_loss: 0.5414 - val_acc: 0.8347
  
  Epoch 00010: val_loss improved from 0.56626 to 0.54143, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 11/20
  
6680/6680 [==============================] - 3s 394us/step - loss: 0.6883 - acc: 0.7820 - val_loss: 0.5493 - val_acc: 0.8467
  
  Epoch 00011: val_loss did not improve from 0.54143
  Epoch 12/20
  
6680/6680 [==============================] - 3s 408us/step - loss: 0.6705 - acc: 0.7859 - val_loss: 0.5410 - val_acc: 0.8359
  
  Epoch 00012: val_loss improved from 0.54143 to 0.54096, saving model to saved_models/weights.best.InceptionV3.hdf5
  Epoch 13/20
  
6680/6680 [==============================] - 3s 412us/step - loss: 0.6338 - acc: 0.7886 - val_loss: 0.5475 - val_acc: 0.8263
  
  Epoch 00013: val_loss did not improve from 0.54096
  Epoch 14/20
  
6680/6680 [==============================] - 3s 404us/step - loss: 0.6152 - acc: 0.8027 - val_loss: 0.5516 - val_acc: 0.8359
  
  Epoch 00014: val_loss did not improve from 0.54096
  Epoch 15/20
  
6680/6680 [==============================] - 3s 402us/step - loss: 0.6085 - acc: 0.8043 - val_loss: 0.5510 - val_acc: 0.8311
  
  Epoch 00015: val_loss did not improve from 0.54096
  Epoch 16/20
  
6680/6680 [==============================] - 3s 403us/step - loss: 0.5809 - acc: 0.8127 - val_loss: 0.5487 - val_acc: 0.8395
  
  Epoch 00016: val_loss did not improve from 0.54096
  Epoch 17/20
  
6680/6680 [==============================] - 3s 400us/step - loss: 0.5627 - acc: 0.8181 - val_loss: 0.5517 - val_acc: 0.8407
  
  Epoch 00017: val_loss did not improve from 0.54096
  Epoch 18/20
  
6680/6680 [==============================] - 3s 415us/step - loss: 0.5507 - acc: 0.8145 - val_loss: 0.5683 - val_acc: 0.8407
  
  Epoch 00018: val_loss did not improve from 0.54096
  Epoch 19/20
  
6680/6680 [==============================] - 3s 416us/step - loss: 0.5325 - acc: 0.8290 - val_loss: 0.5660 - val_acc: 0.8323
  
  Epoch 00019: val_loss did not improve from 0.54096
  Epoch 20/20
  
6680/6680 [==============================] - 3s 410us/step - loss: 0.5174 - acc: 0.8281 - val_loss: 0.5871 - val_acc: 0.8299
  
  Epoch 00020: val_loss did not improve from 0.54096
  

(IMPLEMENTATION) Load the Model with the Best Validation Loss¶

In [93]:
### TODO: Load the model weights with the best validation loss.
  inceptionV3_model.load_weights('saved_models/weights.best.InceptionV3.hdf5')
  

(IMPLEMENTATION) Test the Model¶

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 60%.

In [94]:
### TODO: Calculate classification accuracy on the test dataset.
  inceptionV3_predictions = [np.argmax(inceptionV3_model.predict(np.expand_dims(feature, axis=0))) 
                       for feature in test_inceptionV3]
  
  # report test accuracy
  test_accuracy = 100*np.sum(np.array(inceptionV3_predictions)==
                             np.argmax(test_targets, axis=1))/len(inceptionV3_predictions)
  print('\nTest accuracy: %.4f%%' % test_accuracy)
  
Test accuracy: 80.1435%
  

Sai: There are 2 graphs¶

In [95]:
show_train_history(inceptionV3_model_train_history, 'acc', 'val_acc')
  show_train_history(inceptionV3_model_train_history, 'loss', 'val_loss')
  

(IMPLEMENTATION) Predict Dog Breed with the Model¶

Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan_hound, etc) that is predicted by your model.

Similar to the analogous function in Step 5, your function should have three steps:

  1. Extract the bottleneck features corresponding to the chosen CNN model.
  2. Supply the bottleneck features as input to the model to return the predicted vector. Note that the argmax of this prediction vector gives the index of the predicted dog breed.
  3. Use the dog_names array defined in Step 0 of this notebook to return the corresponding breed.

The functions to extract the bottleneck features can be found in extract_bottleneck_features.py, and they have been imported in an earlier code cell. To obtain the bottleneck features corresponding to your chosen CNN architecture, you need to use the function

extract_{network}
  
  

where {network}, in the above filename, should be one of VGG19, Resnet50, InceptionV3, or Xception.

In [96]:
### TODO: Write a function that takes a path to an image as input
  ### and returns the dog breed that is predicted by the model.
  from extract_bottleneck_features import *
  
  def InceptionV3_predict_breed(img_path):
      # extract bottleneck features
      bottleneck_feature = extract_InceptionV3(path_to_tensor(img_path))
      # obtain predicted vector
      predicted_vector = inceptionV3_model.predict(bottleneck_feature)
      # return dog breed that is predicted by the model
      return dog_names[np.argmax(predicted_vector)]
  

Step 6: Write your Algorithm¶

Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,

  • if a dog is detected in the image, return the predicted breed.
  • if a human is detected in the image, return the resembling dog breed.
  • if neither is detected in the image, provide output that indicates an error.

You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and dog_detector functions developed above. You are required to use your CNN from Step 5 to predict dog breed.

Some sample output for our algorithm is provided below, but feel free to design your own user experience!

Sample Human Output

(IMPLEMENTATION) Write your Algorithm¶

In [97]:
### TODO: Write your algorithm.
  ### Feel free to use as many code cells as needed.
  class DogApp:
      
      def __init__(self,imgPath):
          self.isDog = False
          self.isHuman = False
          self.neither = False
          self.imgPath = imgPath
          self.dogImgpath = "dogImages/train"
          
      def checkDogorHuman(self):
          imgPath = self.imgPath
          if (dog_detector(imgPath)):
              self.isDog = True
          elif (face_detector(imgPath)):
              self.isHuman = True
          else:
              self.neither = True
          
      def displayImage(self, breed):
          
          plt.figure(figsize=(20,10)) 
          
          # A 3-digit integer or three separate integers describing the position of the subplot. 
          # If the three integers are nrows, ncols, and index in order,
          plt.subplot(221)
          plt.title("Testing Image: \n {}".format(self.imgPath.split('/')[2]))
          img = cv2.imread(self.imgPath)
          cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
          imgplot = plt.imshow(cv_rgb)
          plt.subplot(222)
          plt.title("Predicted Dog Breed: \n {}".format(breed))
          dogImgPath = self.getFilePath(breed)
          digImg = cv2.imread(dogImgPath)
          cv_rgb = cv2.cvtColor(digImg, cv2.COLOR_BGR2RGB)
          dogImgPlot = plt.imshow(cv_rgb)
          plt.show()
          
      def getFilePath(self, breed):
          import os
          dirArray = os.listdir(self.dogImgpath)
          for name in dirArray:
              if(breed == name.split('.')[1]):
                  imgDir = self.dogImgpath+"/"+name
                  imgPath = imgDir+"/"+os.listdir(imgDir)[0] # get's the first file from the directory
                  return imgPath
          
      def runApp(self):
          imgPath = self.imgPath
          if(self.isDog):
              breed = InceptionV3_predict_breed(imgPath)
              print("Detected a Dog of Breed {}".format(breed))
              self.displayImage(breed)
          if(self.isHuman):
              breed = InceptionV3_predict_breed(imgPath)
              print("Detected a Human Face, You look like a ... {}".format(breed))
              self.displayImage(breed)
          if(self.neither):
              breed = InceptionV3_predict_breed(imgPath)
              print("Neither Dog nor Human Detected...")
              self.displayImage(breed)
  

Step 7: Test Your Algorithm¶

In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?

(IMPLEMENTATION) Test Your Algorithm on Sample Images!¶

Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.

Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.

Answer:

In [79]:
## TODO: Execute your algorithm from Step 6 on
  ## at least 6 images on your computer.
  ## Feel free to use as many code cells as needed.
  dog_app = DogApp('dogImages/test/130.Welsh_springer_spaniel/Welsh_springer_spaniel_08214.jpg')
  dog_app.checkDogorHuman()
  dog_app.runApp()
  
Detected a Dog of Breed Welsh_springer_spaniel
  
In [170]:
dog_app = DogApp('dogImages/test/048.Chihuahua/Chihuahua_03460.jpg')
  dog_app.checkDogorHuman()
  dog_app.runApp()
  
Detected a Dog of Breed Chihuahua
  
In [98]:
for test_img in glob("images/test/*"):
      app = DogApp(test_img)
      app.checkDogorHuman()
      app.runApp()
      print("\n")
  
Neither Dog nor Human Detected...
  
  Detected a Dog of Breed Petit_basset_griffon_vendeen
  
  Detected a Human Face, You look like a ... Dachshund
  
  Neither Dog nor Human Detected...
  
  Neither Dog nor Human Detected...